From Scent to Scene: Unveiling the Generation of Olfactory Mental Imagery via LLM-Driven Text Mining

Method: This study explores the relationship between perfume scents and consumer imagery associations. Based on the analysis of online reviews, a four-dimensional framework was constructed to decode the patterns of imagery associations in perfume reviews. The research adopts a mixed-methods approach, including text mining, content analysis, multiple regression and fsQCA validation, as well as semantic network mapping. Open Practice Statement: This study only involves text data analysis, poses no ethical risks, and therefore does not require pre-registration. Key Findings: 1) natural materials and cultural symbols significantly enhance product ratings (β=0.0114 and β=0.0132), yet cultural symbols negatively affect popularity (β=-0.0296); 2) fsQCA identifies two high-rating pathways, highlighting the synergistic effects between natural materials and cultural symbols; 3) semantic network analysis reveals distinctive imagery sets for twelve fragrance families and constructs a “ingredient–imagery–scene” mapping chain. Significance: Innovatively, this study proposes a four-dimensional model to quantify the mechanisms underlying olfactory mental imagery generation, filling a theoretical gap in sensory marketing and neuroaesthetics. It also provides practical tools for fragrance brands to leverage dual-driven storytelling through culture and emotion, enabling scene-based customization in response to emerging market trends. Furthermore, the study demonstrates how generative AI can transform unstructured consumer feedback into actionable insights, opening new avenues for consumer experience research in the digital era. In summary, this work offers essential theoretical tools for understanding the core logic of the olfactory economy and provides practical guidance for perfume brands to pursue strategic differentiation.